Finding hidden patterns and root causes of infrastructure failures

Analysis of failure data using a purpose-built system of statistical models combined with LLM assistance. No infrastructure access required. Security-first by design.

↓ View sample report

Built for teams that manage infrastructure for others

IT Service Provider / MSP

You manage IT for multiple clients. Every incident needs a deliverable.

Instead of 3 hours of post-mortem — export the log, share the report. Your client sees structured analysis with numbers, not a summary email.

Internal IT / DevOps Team

Your team closes tickets. The same ones keep coming back.

We show which 3 incident types drive 80% of your problems. Fix those — and the repeating pattern stops.

Growing company, no SRE yet

Reliability questions are coming — from clients, investors, or auditors.

No dedicated reliability engineer? Get a structured picture of your risks without hiring one.

How It Works

1

Export your logs

Any format works. Use whatever you already export from your monitoring system.

CSV JSON TXT XLSX Zabbix Prometheus Grafana
2

Send the file

Upload via the form below and enter your email, or send directly to the Telegram bot. No registration, no installation, no access to your infrastructure required.

3

Get your PDF report

Depending on complexity, within 10 minutes to 48 hours you receive a structured report with findings, risk points, and recommendations — ready to share with your team or client.

What standard monitoring misses

What you see
What our model finds
47 tickets this month. 3 caused real outages. The others — "we'll get to it." Everyone suspects which ones will come back. Nobody has proof.

Another week, another fire drill. Your team is patching symptoms. The cause is still running.

3 clusters. 1 root cause each.

K-Means groups your 47 incidents into 3 behavioral patterns. Three focused fixes instead of forty-seven.

"Things got worse after Tuesday's release." The deployment team disagrees. The meeting goes in circles.

A feeling, not a fact. Hard to escalate without a number.

+44.5% incident rate. Statistically confirmed.

Mann-Whitney U test. p < 0.001. Cliff's δ = +0.45. Now you have a fact to bring to the table.

Same failure type — every week. You fix it, it comes back. The dashboard shows nothing unusual.

You feel the pattern. You can't prove it. The team calls it bad luck.

Every 47 hours. Wednesday at 2 PM. FAP < 0.0001.

Lomb-Scargle detects the hidden cycle with near-zero false-alarm probability. A cron job, a log rotation, a scheduled batch. Fix it once at the root.

Every major outage — the same warning signal appeared 5 minutes before. You only see it in the post-mortem.

You always find the leading indicator after the fact. By then, the damage is done.

database_timeout → app_crash in 89% of cases. Alert before the fall.

Apriori-like lift analysis: lift = 4.2, support = 8.1%, n = 43 event pairs. Set the alert on the leading signal — not on the crash itself.

Numbers from a real dataset

Backblaze Storage Fleet · 1,057 incidents · 89 days · Q4 2023.

217×
Co-failure lift

ST8000NM0055 and TOSHIBA MG08ACA16TA fail together in 85% of cases. Apriori lift = 217 — not coincidence. Shared rack or storage pod dependency found.

−17%
Failure rate drop after change

After December 1: median daily failure rate dropped ~17%. Mann-Whitney U, p = 0.033, Cliff's δ = −0.28 (small but significant). Retiring high-failure drive batches — confirmed by statistics.

94%
Probability of 6h+ downtime

94% probability that any drive failure lasts 6+ hours. GPD fit on 105 tail events, 95% CI confirms. Critical input for SLA and redundancy planning.

2
Root cause clusters

K-Means + silhouette = 0.484. 1,057 drive failures grouped into 2 patterns by failure gap and timing. Two different root causes — two separate remediation tracks.

All calculations are performed by our custom statistical engine. AI assists with interpretation and validation — it generates no figures of its own.

What the report contains

Executive Summary
  • What happened, in plain language
  • Priority ranking (Pareto): which incidents drive most downtime
  • Top 3 recommendations with effort estimates
  • Топ-3 узких места с конкретными таймстемпами
  • Доля простоя от топ-инцидентов (коэффициент Gini)
  • Найденные циклы повторов с привязкой к дню недели и часу
  • Ключевые цепочки «предвестник → отказ» и окно срабатывания
  • Сравнение «до/после» по релизной или организационной границе
  • Прогноз вероятности тяжёлого outage в ближайшие 30 дней (Granular)
  • Что не удалось проанализировать и почему — честность по данным
Technical Details + Methodology
  • Top 5 risk points with statistical evidence (p-value, CI)
  • Failure periodicity chart (Lomb-Scargle)
  • Incident cluster scatter plot (K-Means)
  • Lorenz curve (Pareto / Gini)
  • Boxplot «до/после» с Cliff's δ и доверительным интервалом по bootstrap
  • Таблица ассоциаций событий: lift, support, n — связки «предвестник → отказ»
  • Прогноз хвостовых рисков (Generalized Pareto Distribution, тариф Granular)
  • Раздел «Методология»: используемые методы, α = 0.05, границы применимости
  • Раздел «Ограничения выводов»: где данных не хватило, какие методы отключены
  • Приложение: список таймстемпов топ-инцидентов для разбора
  • Цитирование метода под каждым утверждением — для аудита и CAB
No dashboards. No integrations. Just a file and a structured report.

View a sample real-world audit

Here is a real example — one complete audit on the Backblaze public hard drive dataset (Q4 2023).

Report language:
Download sample report

Upload your log file.
Get the PDF by email

CSV, JSON, TXT, XLSX · up to 20 MB · no registration required

📂
Drag & drop your log file here
CSV · JSON · TXT · XLSX · max 20 MB
✓ File received. Your report will arrive by email in a few minutes.
Something went wrong. Please try again or use the Telegram bot.
or
✈️
Use Telegram bot Send the file directly in the chat — no email needed

Simple pricing. One-time, per audit

Free
$0
free · 1 per email per week
  • MTBF / MTTR analysis
  • Failure periodicity — 1 finding
  • Fact-only AI summary (no recommendations)
  • 1-page PDF report
  • No credit card required
  • AI recommendations
  • Charts & clustering
  • Before/after comparison
Granular
From $600
one-time · per audit
  • Everything in Base
  • Tail-risk forecast (GPD)
  • Event association analysis
  • Dual-model verification (Claude Opus + Sonnet)
  • Extended PDF + raw-data appendix
Подписка
По договорённости
регулярные аудиты
  • Регулярные аудиты по согласованному графику (еженедельно / ежемесячно / по релизам)
  • Гибкий состав методов под вашу инфраструктуру и цели
  • Отдельные условия по объёму данных и SLA на отчёт
  • Напишите нам: hello@opslab.consulting

Free tier: 1 audit per email per week. No credit card required.

Your data stays yours

1

Deleted after 24 hours

Files are permanently removed after your report is delivered.

2

Never used for training

Your data is not used to train any AI model.

3

Auto-masking

IPs, hostnames, and emails are masked automatically before analysis.

4

Без доступа к вашей инфраструктуре

Никаких агентов, VPN-туннелей и доступов к серверам. Вы присылаете только тот файл, который сами решили отправить.

5

Изолированная обработка

Каждый файл анализируется в изолированной сессии. Между клиентами данные не пересекаются и не накапливаются.

Ready to see what's inside your logs?

Upload a file and get the PDF by email — or use the Telegram bot. No commitment. No setup. No access to your systems.